tensorflow/models · error · NotImplementedError

Padded decoding on CPU/GPUs is not supported.

Error message

Padded decoding on CPU/GPUs is not supported.

What it means

Error "Padded decoding on CPU/GPUs is not supported." thrown in tensorflow/models.

Source

Thrown at official/legacy/transformer/transformer.py:128

      sequence. float tensor with shape [batch_size, target_length, vocab_size]
      If target is none, then generate output sequence one token at a time.
        returns a dictionary {
          outputs: int tensor with shape [batch_size, decoded_length]
          scores: float tensor with shape [batch_size]}
      Even when float16 is used, the output tensor(s) are always float32.

    Raises:
      NotImplementedError: If try to use padded decode method on CPU/GPUs.
    """
    inputs = inputs if isinstance(inputs, list) else [inputs]
    if len(inputs) == 2:
      inputs, targets = inputs[0], inputs[1]
    else:
      # Decoding path.
      inputs, targets = inputs[0], None
      if self.params["padded_decode"]:
        if not self.params["num_replicas"]:
          raise NotImplementedError(
              "Padded decoding on CPU/GPUs is not supported.")
        decode_batch_size = int(self.params["decode_batch_size"] /
                                self.params["num_replicas"])
        inputs.set_shape([decode_batch_size, self.params["decode_max_length"]])

    # Variance scaling is used here because it seems to work in many problems.
    # Other reasonable initializers may also work just as well.
    with tf.name_scope("Transformer"):
      # Calculate attention bias for encoder self-attention and decoder
      # multi-headed attention layers.
      attention_bias = model_utils.get_padding_bias(inputs)

      # Run the inputs through the encoder layer to map the symbol
      # representations to continuous representations.
      encoder_outputs = self.encode(inputs, attention_bias, training)
      # Generate output sequence if targets is None, or return logits if target
      # sequence is known.
      if targets is None:

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/legacy/transformer/transformer.py:128 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/d96a3d7ad0acf9f8. Report an issue: GitHub.